Messages keep getting clearer
Rules for stories and science
George Saunders is a celebrated author known for his short, often humorous, stories. He started out as an engineer, though, and kept an analytical bent. In 2021’s A Swim in a Pond in the Rain, Saunders breaks down short works by Russian masters like Chekhov and Tolstoy to draw out “universal laws of fiction.”
David Deutsch is an Oxford physicist known for seminal work in computation theory. When not propounding the many-worlds interpretation of quantum mechanics, he moonlights in philosophy of science. In The Beginning of Infinity, Deutsch argues that scientific progress consists of creating better and better explanations for the world, going into detail on what this means.
These works have an unexpected harmony. While stories and science are often put in opposition, the qualities of a good narrative and a good scientific theory are fundamentally similar. I’ll first review each author’s ideas, then try my own explanation for their likeness.
George Saunders’ Universal Laws of Fiction
Be specific
Detail makes characters and worlds believable. Die Hard takes place on Christmas Eve, not a random day in March. Specific details, like props in a play, are there to perform “story-specific work.” The fact that it’s Christmas is why McClane comes to LA, why the office workers are still at Nakatomi Plaza, and why John is trying so hard to reconcile with Holly. Good writing involves “continually revising toward specificity, so that specificity can appear and then produce plot."
Honor efficiency
Insignificant details annoy us; specificity must serve a purpose. Good stories are “ruthlessly efficient…nothing exists in a story by chance.” Chunk and Sloth bond over a Baby Ruth, not a generic piece of candy. That Baby Ruth starts the child-like Sloth on a developmental arc away from the Fratellis and toward the Goonies. In The Usual Suspects, stray comments imply that Fenster and McManus are lovers; revenge for Fenster’s murder later motivates McManus to join the final heist. Details should matter!
Always be Escalating
Like a 9-year-old boy recounting Infinity War, a straight series of events is boring AF. What’s missing is causation: each beat impelling the next. Causation creates meaning and energy, turning a sequence into a “continual system of escalation” headed somewhere unexpected. In After Hours, Griffin Dunne just wants to meet a nice girl, but ten hours later he’s encased in plaster, hiding from a vigilante mob. In Something Wild, Jeff Daniels just wants to, well, meet a nice girl, but a few days later he’s in a deathmatch with Ray Liotta. Saunders references an apocryphal Einstein quote: "no worthy problem is ever solved in the plane of its original conception."
David Deutsch on Good Explanations
“What made science effective at understanding the physical world when all previous ways had failed?” Deutsch’s answer is that science, and other knowledge, relies on creating better and better explanations for the world. We don’t rely on induction or experimentation or derivation from first principles to do this; creating a new explanation is “inherently an act of creativity.” People adapt, remix and mashup old ideas to conjecture new ones, which are then criticized and tested. Most new ideas don’t survive, a process akin to evolution by natural selection.
A scientific theory explains some aspect of reality; other hallmarks like prediction and testability follow from this. When we watch a stage magician perform a trick, sawing his assistant in half or bringing up an onlooker to pick a card, it’s easy to make an accurate prediction: the assistant will later appear unharmed, the magician will pull the eight of clubs from his back pocket. But no one really cares. We wonder how it works.
Most explanations are bad
Each year, the Greek goddess Persephone is periodically confined to the underworld, with the resultant mourning of Demeter bringing on winter. This theory indeed makes testable predictions: if Persephone escapes, spring will come early; winter should occur everywhere at the same time; if enough sacrifices are made to Demeter, winter will be mild. The actual myth contains a wealth of detail, but none of it matters. Each part of the story could change, and it would still explain the seasons just as well, just as every culture had its own equally useless myths.
Good explanations are “hard to vary”
On the other hand, an explanation that is hard to vary is both specific and causal. “It will rain at 4pm because the warm front 200 miles away moving in at 10mph from the southwest will meet the low pressure system already in the area ” is better than “It will rain later because the sky was red this morning.” Specificity makes the theory more useful, and the explanation is hard to vary because changing one detail (the location of the warm front) necessarily affects others.
Theories with superfluous parts are bad because those details can change with no effect on the rest. Newton could have argued that objects attract each other with a force proportional to their masses and inversely proportional to the square of the distance, all divided by a leprechaun factor. This makes a worse theory because it’s unnecessary, and why not dragons or King Arthur? An easily varied detail does no work, so good theories tend toward fewer moving parts.
Good explanations have “reach”
A compelling account of one data set answers other, unexpected questions. Evolution by natural selection explains the fossil record and the front page of Reddit. Once we accept that the Earth’s axis-tilt explains why Paris has four distinct seasons, we know that Boston should have similar seasons, Sri Lanka not so much, that weather in Pretoria should differ from Oslo, and that Mars and Saturn have seasons of their own.
A science story: this ain’t funny so don’t you dare laugh
Myths, fairy tales, novels, explanations, and scientific theories are all descriptions of reality, or a reality; they tell us how a situation came to be. Put differently, both stories and explanations are messages, transmitted from one brain to another.
Messages contain information, both the intended signal and accompanying noise. The problem of sending messages over a noisy channel was formalized in 1948, when mathematician Claude Shannon birthed the field of Information Theory, Athena-like, from his skull. Its’ concepts apply to all forms of communication, not just guitar amps and data compression. Information Theory unites Saunders’ tips on writing and Deutsch’s take on science, illustrating universal principles at work when we talk, write, and think.
Information reduces uncertainty
In a conversation or text or essay, the number of things a sender could say is vast, and the receiver doesn’t know what to expect. The more possibilities, the higher the “entropy.1” Information in the signal resolves your uncertainty: it’s this particular message. Without uncertainty, there’s no information, because the point of communication is telling people something they don’t yet know.
Imagine a text from your mom: “Check out the front page of the Times!” You’re expecting a news article, but it could be almost anything: a celebrity died, cold fusion works, the Mounties have taken Sault Ste. Marie. Some messages are more likely than others, but you must read the article to know for sure. The more surprising the news, the more information you gain.
We like stories, and explanations, with a lot of information. Specific details reduce uncertainty about the world we’re visiting: is this grizzled detective the type with a house full of jazz LPs or the type who feeds stray tabbies? Thinly drawn characters and settings don’t resolve much uncertainty. In the same way, hand-wavy explanations don’t cut it: “seasons happen because the earth moves around” leaves a lot on the table. Specificity is information, telling us exactly which message we’re receiving, and exactly which world we’re in, out of all the options.
We also like surprise because it reveals greater information, an unlikely message. This could be the last-minute twist of Primal Fear, Detective Columbo laying out how the killer did it, or plate tectonics explaining how the Scottish Highlands, the Appalachians, and the Atlas Mountains of Morocco were once the same range. Unexpected insights have more impact.
A difference that makes a difference2
Larding up a story with detail only goes so far. It’s the classic blunder of literary fiction and hard sci-fi alike: tedious details that don’t matter and are thus easy to vary. These are not signal but noise, a background hiss. If the drapes were sheer instead of linen, would it affect anything else? Does the starship drive running on antimatter make a difference to our characters? Pointless detail - that which is easily varied - does no work.
Much like humans, stories need structure
Three-hundred pages of gibberish is specific and surprising, in its way, but it’s also indistinguishable from random noise. Language conventions enable meaning while shrinking the space of possibilities. When every Q must be followed by a U, the system has fewer degrees of freedom, lower entropy, and greater structure.
As with spelling and grammar, storytelling rules constrain the message by connecting elements and imposing logic. A good story needs consistent characterization, a sensible progression of events, a narrative arc. Surprise is delightful when earned, like The Sixth Sense or Signs. Endings out of left field leave a bad taste, like every other Shyamalan film.
Take Back to the Future, Galaxy Quest, or The Terminator. Those plots fit together like a goddamn Swiss watch! The satisfaction of such interlocking stories, with call-backs, pay-offs, and everything coming together, is key. Characters, setting, and plot events don’t just bounce around, they form bonds that are hard to break. The stronger the connections, the clearer the signal.
Scientific explanations face a greater constraint: they must match our particular reality, not just some possible reality. This winnows the degrees of freedom, or it should (looking at you, psychoanalysis!) When new data appears, such as galaxies that appear older than the universe, the rest of the story must change. Like Chinatown and Tootsie, good theories have an interlocking quality, with no freely moving parts.
The noise is coming from inside your head!
We picture messages going from one person to another, with noise picked up en route. But for actual humans, most “noise” arises at the source, with fuzzy thinking, errs and umms, and turgid prose.
Putting an idea down, reading it, tweaking it, rinse and repeat. Writing and editing is sending a message over a noisy channel, but the sender and recipient are one. Each round of edits boosts signal and reduces noise: making a point clearer, dropping a digression, revising toward specificity, omitting needless words. Note the parallels to arriving at an elegant scientific theory or simplifying a mathematical expression.
Crafting a good story or explanation takes a lot of work, but that energy is stored in the message. We beaver away, arranging and organizing and tweaking, to lower the entropy of our little system. Like a charged-up battery, the energy is then available to do work on others: leave an emotional impression, alter their perspective, or impart new knowledge.
It’s similar to the entropy you recall from high school. Thermodynamic entropy is a statistical measure of all the possible arrangements of a physical system. Information entropy corresponds to the number and likelihood of all potential messages in a system.
A well-known description of information from Gregory Bateson, 1972.